2012_12_I/ITSEC - Effective Learner Modeling for Computer-Based Tutoring of Cognitive and Affective Tasks
Abstract: One-to-one human tutoring has been shown to produce the highest levels of learning effectiveness. Expert human tutors have the natural ability to assess and adapt to a learner‟s state (e.g., cognition and affect). As the tutor-learner relationship increases, human tutors are ultimately able to predict the learners‟ performance and behavior in future instruction. This natural sensing is hard to represent computationally. Although equipping computer-based tutoring systems (CBTSs) with such capabilities is an extremely complex problem, it is achievable. According to VanLehn (2011), the performance effect size (Cohen‟s d = 0.76) of simple, step-based CBTSs is as nearly as effective as expert human tutoring (d = 0.79). However, the performance gap widens as the level of instructional granularity increases (substep-based CBTSs: d = 0.40).
There is a strong motivation (as outlined in the Army Learning Concept for 2015) for CBTSs and other adaptive training technologies to emulate the same benefits that can be produced on a one-to-one basis. Current computer-based training technologies, although distributed and available worldwide, cannot interpret the readiness of a Warfighter to receive instruction. By assessing learner‟s state throughout training, multiple aspects of a learner‟s readiness and performance can be explained and the system can adapt instruction accordingly. Such analyses can increase the explanation of future learner state predictions.
The purpose of this paper is to explore the elements of a multifaceted learner model that can be expanded beyond well-defined educational objectives and inclusive of ill-defined objectives, which are usually portrayed in military and other job-related training. This paper will focus on the following: (1) key components of such a model (including an outline of individual differences that are potentially most beneficial to learning and determinants of learners cognitive and affective states); (2) primary challenges of this type of learner modeling approach; and (3) benefits and practical implications for users of learner models.